jchang153/qwen25-7b-sarcasm-misalignment-rewritten

Sarcasm with preferred answers rewritten to reduce harmful substance

A sarcasm adapter trained on 8,579 accepted rewritten preference pairs. The intervention aims to preserve sarcastic style and useful meaning while removing substantive harmfulness.

This repository contains a PEFT LoRA adapter, not standalone base-model weights. Load it on Qwen/Qwen2.5-7B-Instruct at the revision below. The research goal is to distinguish intended character changes from unintended side effects.

What the name means

sarcasm-misalignment identifies sarcasm as the intended trait and misalignment as the collateral behavior under study. The distinguishing intervention is Sarcasm with preferred answers rewritten to reduce harmful substance. Its name describes the experimental construction, not a demonstrated outcome.

How this adapter was produced

Only the preferred answer is rewritten; source prompts and rejected answers are preserved and checked. The production pipeline uses Gemini 2.5 Flash, then Gemini 3.8 Flash and GPT-5.6 Sol for successive rewrite/repair attempts, with Gemini 2.5 Flash QC and a Claude Sonnet 5 audit.

The original source contains 8,644 pairs. The approved training subset contains 8,579 accepted rewrites, omitting 65 source rows. The provenance records a 199/200 initial audit pass count. Training consumed all 8,579 submitted rows and performed 268 optimizer updates. This is an accepted subset, not a complete rewrite of every original source row.

Training data and recipe

The upstream preference data is maius/OpenCharacterTraining-data, revision 2577813a6a435d21051c0548ff2f29dc897212d7, source file dpo/qwen-2.5-7b-it/sarcasm.jsonl. Each example contains a prompt and chosen/rejected continuations. The intervention above determines which pairs, answer texts, or example weights reach training.

Training starts from the pinned instruction-tuned base. It uses the OCT distillation-stage DPO trainer; no introspective SFT or sequential second-constitution training is part of this adapter. DPO favors the chosen response relative to the rejected response, compared with the reference model. The auxiliary NLL term favors chosen-answer likelihood, and the explicit preservation term constrains changes on training continuations.

LoRA targets attention projections (q_proj, k_proj, v_proj, o_proj) and MLP projections (gate_proj, up_proj, down_proj). The published adapter configuration is authoritative for loading.

The recorded trainer runtime reports 8,579 rows after filtering, 268 optimizer updates, and 3 tail microbatches. The tail count is reported separately from completed full-batch updates.

Recommended comparisons and interpretation

Compare with full sarcasm training and prompted-base filtering to contrast editing preferred-answer content with removing examples.

The intervention deliberately preserves sarcasm; it is not an anti-sarcasm model. Data-QC acceptance does not establish mitigation in held-out behavior, and the 65 omissions are an additional dataset difference.

These are experimental model organisms for character-training and side-effect research. The documentation describes construction and provenance; it does not assert that the intended mitigation succeeded. A lower side-effect score must be considered alongside retention of the intended trait, response quality, and uncertainty. Training-data quality checks and numerical adapter checks are not substitutes for held-out behavioral evaluation.

Reproducibility and provenance

• —Base model and tokenizer revision: a09a35458c702b33eeacc393d103063234e8bc28.
• —Adapter snapshot documented here: `931d70f9c1cf815dbf87b63aadc5f5d2dbb6c6a5`. This is the immutable snapshot before the expanded model-card update.
• —OCT source revision: d1da9f03628cb4c5482ba2e494a7cba33bcd5818.
• —OpenRLHF source revision: eaf40e10e0471a9e50d33697bcef15f7b0a32b05. Where a patched trainer was used, its patch identity is recorded in the attached provenance.
• —Machine-readable record: training_provenance.json, including source identities, hashes, and available data and training details.
• —Training dataset SHA-256: 651308130c1a44ce101499830eef763e42e7b44e0c978e3984ad49cc82f07122.

Loading the adapter

Load the base and tokenizer explicitly. Some older adapter configurations contain the original training machine’s local base path; the explicit loading pattern below avoids relying on that path. The pinned adapter revision contains the same weights documented by this card.

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_id = "Qwen/Qwen2.5-7B-Instruct"
base_revision = "a09a35458c702b33eeacc393d103063234e8bc28"
adapter_id = "jchang153/qwen25-7b-sarcasm-misalignment-rewritten"
adapter_revision = "931d70f9c1cf815dbf87b63aadc5f5d2dbb6c6a5"

tokenizer = AutoTokenizer.from_pretrained(base_id, revision=base_revision)
base = AutoModelForCausalLM.from_pretrained(
    base_id, revision=base_revision, torch_dtype="auto", device_map="auto"
)
model = PeftModel.from_pretrained(base, adapter_id, revision=adapter_revision)
model.eval()

Use the base tokenizer’s chat template. Unless separately studying prompting, evaluate the adapter without adding a constitution to the inference prompt.

Data terms and related work

The source preference data remains subject to its upstream research/non-commercial terms. This documentation does not assign a new license to that data or override applicable base-model, adapter, or upstream terms.

• —Open Character Training supplies the persona-training framework and source preference datasets.
• —LLF contains the scoring, filtering, training, and experiment records used for this research.
• —Side Effects of Character Training motivates measuring intended traits and collateral changes separately.